New Zen microarchitecture details

[Phynaz]

still waiting :colbert:

your all tip and no iceberg

Waiting for what? Me to join the ADF?

 

[Phynaz]

Look at this blatant trolling.

Look at your blatant mod callout.
(Which I reported BTW)

Anyway, what was a troll about that? PPB stated facts and how they could affect performance of Zen. He was doing well in presenting his view. Then he blew his leg off.

If he would have just left the last sentence off his post...Ah well.

 

[itsmydamnation]

Waiting for what? Me to join the ADF?

Nope to answer the opposite side of question i gave you the courtesy of answering.

 

[Phynaz]

Nope to answer the opposite side of question i gave you the courtesy of answering.

I've already presented my case but you don't want to hear it. There isn't an answer I can give you that doesn't involve me being an AMD fanboy that will satisfy you. Upon examining your post history I can see why.

 

[MajinCry]

Phynaz, you've not provided anything to show that Zen is going to be this lackluster failure. All you've done is pointed at Bulldozer and proclaimed that Zen is of the same fate.

This ain't hard to understand. Substantiate your claims, just like the other lads have, by discussing shortcomings/improvements in regards to the architecture.

 

[Arachnotronic]

Phynaz, you've not provided anything to show that Zen is going to be this lackluster failure. All you've done is pointed at Bulldozer and proclaimed that Zen is of the same fate.

This ain't hard to understand. Substantiate your claims, just like the other lads have, by discussing shortcomings/improvements in regards to the architecture.

Not much is actually known about the architecture, but plenty is known about the company developing it & its track record. I know it's hard for "tech nerds" to try to view things from the perspective of, say, managers, but it's an important perspective and one that should not be dismissed.

It's all too easy to "imagine" a wunderchip and compare it to a chip that's actually available and conclude that said wunderchip is the best thing ever, but there are business, financial, and resource realities that need to be taken into consideration as well.

 

[Phynaz]

^Truth.

 

[RussianSensation]

Why are you comparing Zen's 40% IPC increase to Piledriver? The 40% increase is over Excavator, it's process independent, and AMD has been saying lately it's more likely 45% now.

Unless AMD is lying out their teeth, Zen should be WAY less than 10% behind Skylake...

What's the real world IPC increase of Excavator over Piledriver, 10-15% max? Big deal.

I just pitted a 4.7-5Ghz Piledriver against a 4-4.2Gh Skylake; and it got smashed by 82%. Add Skylake OCing, its single core performance leads by 90%.

Even if you add 40% IPC, add another 20% for Excavator over Piledriver, you are still miles away from Skylake. And that assumes Zen will Turbo to 4.7-5Ghz. It won't since high IPC architectures do not clock high as easily. If Zen launches with 4-4.2Ghz Turbo, that alone accounts for the IPC differential between FX9590 and Excavator. Other people already pointed this out. If you re-read my post in detail and consider the context, it's impossible for AMD to make a 95W chip with similar single threaded performance as Skylake.

Also, do I need to remind people in this thread that we will have i7 6800K-6950X and i7-7700K before Zen drops? It's funny how Zen launches in late 2016 so we should compare it to 5820K/6700K? I might as well make the argument straight up that if Zen volume production --> retail is only Q1 2017, we actually should be comparing it to the rumored Q2 SKL-E and Q3 2017 Cannonlake. What kind of a double standard is it that we use 'old' Intel CPUs/architectures but Zen launches way later?

Right now I can go out and buy an i7 6700K and enjoy it until December 1st. I bet Zen won't even be out in large volumes by then. It's like this forum assigns 0 value for opportunity cost of waiting but at the same time the minute Zen launches (and we were supposed to wait for it for 15+ months since August 2015 I7 6700K launched), we are supposed to ignore that Cannonlake and SKL-E won't be far away either?!

As I said, even with a ~ 10% IPC differential between 5820K and 6700K, 6700K still sells like hot cakes. I think too many people here underestimate that many consumers will still choose 4 fastest cores over 6-8 slower ones. I would likely take i7-6800K BW-E over i7 6700K but I would take i7-6700K over i7 4930K. If Zen brings 6-8 cores with IVB or even Haswell IPC, even that isn't a slam dunk. Even ignoring i7-7700K (Kaby Lake), we will have $389 i7-6800K BW-E probably in June/July of this year.

It's also interesting how people believe that AMD will by pure magic and a fraction of R&D make up 4 (!) major Intel architectures:

Nehalem
Sandy
Haswell
Skylake

There is no way that AMD can come up with an architecture that incorporates a boost in single threaded performance = to 4 of Intel's major architectures since 2008. It took Intel 7 years but AMD can short-cut it? Ya right...

You guys are literally setting up Zen to fail by having insanely unrealistic expectations. Even Intel's 6-core 3.6Ghz BW-E will have a 140W TDP and Intel's 14nm tech guaranteed wipes the floor with GloFo's FinFET. Yet, people only look at the architectural aspect and ignore that Intel's 14nm node is no way comparable to their competitors -- it's far superior.

And here is the kicker -- even if by pure magic, AMD can pull out an 8-core Zen that matches i7-5960X, Intel will just add more cores and drop prices on SKL-E to make sure AMD is not relevant. Why? Because Intel has 60%+ gross margins and their BW-E workstation platform already has a 22/44HT CPU. Let's not kid ourselves -- Intel is the one who decides just how high AMD can price Zen. The minute Intel feels threatened, they'll move 6-core SKL-E go $299.

AMD's best bet is to win OEMs/mobile designs. As far as DIY market goes, I don't think it has a chance because Intel can drop prices/raise the number of cores at will.

 

[deasd]

Not much is actually known about the architecture

You really don't believe, don't understand, or even haven't read this thread's poster, dresdenboy's study, right?

Track record? What about Netburst? Larrabee? Itanium? What about NVidia's Fermi? Well, I understand people who don't know anything about architecture and programming can only cite a company's track record to 'prove' something.

(I don't wanna see this thread being closed, as long as there's still some valuable posts here.)

 

[EightySix Four]

What's the real world IPC increase of Excavator over Piledriver, 10-15% max? Big deal.

I just pitted a 4.7-5Ghz Piledriver against a 4-4.2Gh Skylake; and it got smashed by 82%. Add Skylake OCing, its single core performance leads by 90%.

Even if you add 40% IPC, add another 20% for Excavator over Piledriver, you are still miles away from Skylake. And that assumes Zen will Turbo to 4.7-5Ghz. It won't since high IPC architectures do not clock high as easily. If Zen launches with 4-4.2Ghz Turbo, that alone accounts for the IPC differential between FX9590 and Excavator. Other people already pointed this out. If you re-read my post in detail and consider the context, it's impossible for AMD to make a 95W chip with similar single threaded performance as Skylake.

Also, do I need to remind people in this thread that we will have i7 6800K-6950X and i7-7700K before Zen drops? It's funny how Zen launches in late 2016 so we should compare it to 5820K/6700K? I might as well make the argument straight up that if Zen volume production --> retail is only Q1 2017, we actually should be comparing it to the rumored Q2 SKL-E and Q3 2017 Cannonlake. What kind of a double standard is it that we use 'old' Intel CPUs/architectures but Zen launches way later?

Right now I can go out and buy an i7 6700K and enjoy it until December 1st. I bet Zen won't even be out in large volumes by then. It's like this forum assigns 0 value for opportunity cost of waiting but at the same time the minute Zen launches (and we were supposed to wait for it for 15+ months since August 2015 I7 6700K launched), we are supposed to ignore that Cannonlake and SKL-E won't be far away either?!

As I said, even with a ~ 10% IPC differential between 5820K and 6700K, 6700K still sells like hot cakes. I think too many people here underestimate that many consumers will still choose 4 fastest cores over 6-8 slower ones. I would likely take i7-6800K BW-E over i7 6700K but I would take i7-6700K over i7 4930K. If Zen brings 6-8 cores with IVB or even Haswell IPC, even that isn't a slam dunk. Even ignoring i7-7700K (Kaby Lake), we will have $389 i7-6800K BW-E probably in June/July of this year.

It's also interesting how people believe that AMD will by pure magic and a fraction of R&D make up 4 (!) major Intel architectures:

Nehalem
Sandy
Haswell
Skylake

There is no way that AMD can come up with an architecture that incorporates a boost in single threaded performance = to 4 of Intel's major architectures since 2008. It took Intel 7 years but AMD can short-cut it? Ya right...

You guys are literally setting up Zen to fail by having insanely unrealistic expectations. Even Intel's 6-core 3.6Ghz BW-E will have a 140W TDP and Intel's 14nm tech guaranteed wipes the floor with GloFo's FinFET. Yet, people only look at the architectural aspect and ignore that Intel's 14nm node is no way comparable to their competitors -- it's far superior.

I agree with you that there is almost no way Zen will meet some of the people in this thread's expectations, but I disagree a bit about being unable to catch-up. A lot of the improvements of those 4 major architectures were enabled by smaller processes, which AMD is about to get a lot closer to Intel on and this doesn't take into account that AMD can learn from Intel's successes in that time frame. Neither of these companies operate in a vacuum and AMD can leverage the knowledge gained by Intel in the meantime simply by learning from their hardware. Executing on the knowledge is a whole different ball game however, so I have comparatively lower expectations.

AMD will close the gap and may lay the foundation to accelerate their hardware development to further close the gap, but they'll still be behind. The question is, can AMD make a logical value proposition for purchasing their hardware instead - if it's not faster it needs to be better at something. Perf/$, Perf/Watt, Perf/Rack, etc.

 

[RussianSensation]

I agree with you that there is almost no way Zen will meet some of the people in this thread's expectations, but I disagree a bit about being unable to catch-up. A lot of the improvements of those 4 major architectures were enabled by smaller processes, which AMD is about to get a lot closer to Intel on and this doesn't take into account that AMD can learn from Intel's successes in that time frame. Neither of these companies operate in a vacuum and AMD can leverage the knowledge gained by Intel in the meantime simply by learning from their hardware. Executing on the knowledge is a whole different ball game however, so I have low expectations.

You know Intel has moved to Process-Architecture-Improvement now? Well, SKL-E will be manufactured on 2017 14nm tech. That means Intel will have the knowledge and node maturity/yield of Broadwell, Skylake, Broadwell-E and Kaby Lake. AMD has to release a new architecture on an unproven, immature FinFET node. Think about it, if you could port SKL-E to GloFo's FinFET, Intel's 14nm SKL-E would beat it in die size and perf/watt / leakage. That's my point too -- people here are only looking at the architectural side while straight up ignoring just how mature Intel's 14nm will be by early 2017.

There is probably a reason AMD is dropping iGPU from desktop Zen -- they know they have to offer more cores than Intel to be able to price them competitively and overcome the single threaded IPC disadvantage. That's probably why they chopped off the idea of having the iGPU to use the extra die space, TDP headroom to add more cores and up the clocks since they know there is no chance their 4-core Zen ~ 4-core Skylake HT.

 

[Sweepr]

Track record? What about Netburst? Larrabee? Itanium? What about NVidia's Fermi? Well, I understand people who don't know anything about architecture and programming can only cite a company's track record to 'prove' something.

History shows us that the so called 'experts' (some of them still here) who mocked anyone who didn't believe in their '3.5-4.0 GHz+ Bulldozer improves IPC while doubling core count at mainstream' predictions were utterly wrong. Also, there's a lot of speculation in the current Zen technical analysis around the web.

 

[RussianSensation]

I don't think Intel is stupid either. They want to keep their profit margins but not bankrupt AMD either. It actually could make sense for Intel to raise prices. Move 10-core 6950X to $1499, keep 8-core i7-6900K at $999. This allows AMD to price 8-core Zen at $499-699. Intel keeps its profit margins and AMD doesn't need to deliver a miracle. I have a feeling people who have been buying $120-150 8-core FX chips expecting an 8-core 16 threaded Zen for the price of an i7-6700K are in for a shock.

Let's follow that pro-AMD logic in this thread. If I had a winner on my hands with Zen, then I am pricing it appropriately. Why the hell would I sell an 8-core for the price of an i5-6600K/6700K?

Too many Zen contradictions in this thread imho. Do people honestly believe if Zen is a fire cracker that AMD will be selling it for $249-349?

 

[majord]

I don't know where people got the "Haswell IPC level" expecation from.. some vague rumor I think.

AMD's own performance estimations haven't changed.. 40% over Exv, you can't translate that to HW IPC, period.

I think everyone should get back to the expectation of Ivy levels on average, potentially lower AVX2. That is not unrealistic, nor is it in any way a bad thing, given the performance uplift from both HW and SL have come at the expensive of Power consumption and die area - the later quite proportionality (about 35+% larger core area node-node going from Ivy to SL)

 

[deasd]

What's the real world IPC increase of Excavator over Piledriver, 10-15% max? Big deal.

There's still too little test around the web but I think Excavator with L3 cache has average 15-20% advantage against Piledriver.

I just pitted a 4.7-5Ghz Piledriver against a 4-4.2Gh Skylake; and it got smashed by 82%. Add Skylake OCing, its single core performance leads by 90%.

It's important to note that Skylake outperform Nehalem by 70% in some FPU-stress test(especially Cinebench R10)
http://www.anandtech.com/bench/product/1646?vs=99

Such huge gap is very rarely seen in most of other benchmarks, also in the same way you compared intel and amd. I can't explain but I believe Intel optimizing vector SIMD very hard since Sandybridge and later gen, such as compilers and vectorizing and enlarging L/S of pipeline. But such these optimization wouldn't take affect for scalar performance(superpi, wprime, Fritzchess), even these changes are degrading efficiency of these test which cannot vectorize(as you can see some gen with wider machine consume much more power when utilizing longer instruction set). Considering Intel has no advantage/willing of pushing low-level heterogeneous computing like AMD/Nvidia did, it's understandable that intel to elaborate its own advantage further. But it is just a disparity between different vendors.

AMD might have some works on this for Zen too, I can't sure. But I think it won't have a huge leap like from Nehalem to Skylake. I guess Zen MIGHT have Sandybridge level performance in these tests.

 

[Doom2pro]

What's the real world IPC increase of Excavator over Piledriver, 10-15% max? Big deal.

IPC is measured relative to a single clock speed... Comparing IPC Piledriver clocked higher to lower clocked IPC of Excavator is meaningless because the IPC is compared with the same clock.

 

[majord]

It's also interesting how people believe that AMD will by pure magic and a fraction of R&D make up 4 (!) major Intel architectures:

Nehalem
Sandy
Haswell
Skylake

There is no way that AMD can come up with an architecture that incorporates a boost in single threaded performance = to 4 of Intel's major architectures since 2008. It took Intel 7 years but AMD can short-cut it? Ya right...

I'm sorry but you seem to not grasp the situation properly if you think this is how it works. They are not simply trying 'catch up' , they've moved back to a wider, high IPC design. There will of course be some elements of evolutionary architectural improvements - deeper buffers, Ld/store improvements , brance prediction improvements, etc, etc, and given the big jump down to 14nm, this will no doubt be significant part of the 40% uplift over Excavator. but the bulk will be from the physically wider execution (integer in particular), lower latency cache system, and (unconfirmed I know) shorter pipeline.

 

[DrMrLordX]

You guys want to see some Excavator vs Vishera action? Here's Cinebench R10:

Excavator (3.4 GHz, Carrizo reference platform, static clockspeed):

http://cdn.overclock.net/0/05/052237f1_CZ-CB10-3.4G-Static.png

Piledriver (3.8 GHz, Vishera from one of Flank3r's reviews):

http://i.imgur.com/uRxJpjA.jpg

In a legacy fp workload utilizing SSE2, 2m Carrizo with no L3 cache scores 13146. 2m Vishera with L3 cache scores 12751 with a 400 MHz clockspeed advantage (not to speak of turbo). The same Vishera should score 11408 @ 3.4 GHz, giving XV an advantage of 15% without the benefit of L3.

 

[Arachnotronic]

I don't think Intel is stupid either. They want to keep their profit margins but not bankrupt AMD either. It actually could make sense for Intel to raise prices. Move 10-core 6950X to $1499, keep 8-core i7-6900K at $999. This allows AMD to price 8-core Zen at $499-699. Intel keeps its profit margins and AMD doesn't need to deliver a miracle.

RS, gross profit margin is a function of product competitiveness. If AMD has an 8-core Zen that's competitive with an 8 core 6900K, then Intel would be downright stupid to price it for $999 if AMD is pricing roughly equivalent product for $499-$699.

Semiconductors are a volume game and in a shrinking PC market throwing away volume like that (because I guarantee you that such gross overpricing would crater unit sales) is dumb.

The real problem for AMD is that Intel has a lot of knobs it can dial in response to AMD product. Intel has 10 core, 15 core, and 24 core Broadwell-EP chips that it can configure any which way it likes & unlock the multiplier in response to what AMD puts out. By the time Zen arrives, we are probably talking more about SKL-EP rather than BDW-EP, anyway.

 

[cytg111]

Not much is actually known about the architecture, but plenty is known about the company developing it & its track record. I know it's hard for "tech nerds" to try to view things from the perspective of, say, managers, but it's an important perspective and one that should not be dismissed.

It's all too easy to "imagine" a wunderchip and compare it to a chip that's actually available and conclude that said wunderchip is the best thing ever, but there are business, financial, and resource realities that need to be taken into consideration as well.

Same company that devised GCN, console wins, engineered mantle to make the next directx superior on their respective hardware? That company? We talked that up a year ago and noone no-one really believed they could pull it off. Where are we now? So yea, smack it up, smack it down. I just know that those in the circle of shame wil be the silly peeps who climbed the trees highest, almost impossible to get down from there.

 

[Erenhardt]

How slow zen will be:
We start with fx8350 base of 6.85 CB11.5 MT multiply it by XV IPC increase 1.1 and announced Zen improvements of 1.4 and we have 10.5 CB11.5 MT score which is more than i7 6700k.

Will Zen have more 'module penalty' than Vishera?

 

[inf64]

How slow zen will be:
We start with fx8350 base of 6.85 CB11.5 MT multiply it by XV IPC increase 1.1 and announced Zen improvements of 1.4 and we have 10.5 CB11.5 MT score which is more than i7 6700k.

Will Zen have more 'module penalty' than Vishera?

There is no module penalty with Zen since there are no modules like in Bulldozer. We will have discrete cores with SMT, so you will have opposite of penalty in MT code: SMT should give ~20-25% boost in MT code, similar to what i7 gets (if AMD did SMT right).

 

[cytg111]

How slow zen will be:
We start with fx8350 base of 6.85 CB11.5 MT multiply it by XV IPC increase 1.1 and announced Zen improvements of 1.4 and we have 10.5 CB11.5 MT score which is more than i7 6700k.

Will Zen have more 'module penalty' than Vishera?

The absolute best guesses towards Zen performance is that offered by dredsenboy and that is withholding whatever glofo adds to the party. In short, noone knows for real, either way.

 

[myocardia]

We start with fx8350 base of 6.85 CB11.5 MT multiply it by XV IPC increase 1.1 and announced Zen improvements of 1.4 and we have 10.5 CB11.5 MT score which is more than i7 6700k.

Yes, as long as the 8 core/16T Zen runs @ 4.0 Ghz base, and 4.2 Ghz turbo frequencies, which I can assure you will not be happening. I'd be surprised if the 16 thread Zen had a base frequency as high as 3.0 Ghz, although that would in fact be a nice surprise! edit: Remember, this is only supposed to be a 95 watt CPU. Divide those 95 watts by 16 threads, and it doesn't allow frequencies of anywhere near Vishera.

Will Zen have more 'module penalty' than Vishera?

Zen will have no modules, so there won't be any module penalty at all. Like Intel's CPUs, it will either gain a decent amount of performance, when the software is able to use more than 8 coresthreads, or it will lose a percentage point or two, when certain software doesn't benefit from SMT, like we see in a few games currently.

 

[majord]

I know the whole CMT vs SMT thing has been done to death over the years, but it seems people still find it hard to get their head around, and it shows as people struggle to make comparisons between Zen and Bulldozer based cores.

using the terms CMT or modular "Penalty", and SMT "advantage" are really just different sides of a coin, describing the same thing, and it comes down to one having more of its resources dedicated to each thread (CMT).

Note the subtle distinction there - more OF it's resources, not necessarily MORE resources. i.e, a BD or Vishera 'module' has less resources than a single SMT Intel core, even going back to Haswell (?) . It may have slightly more execution ports in total if you add them all up - but everything else, including register, buffer sizes Decode etc is all behind or equal.

Kaveri complicates things by duplicating the decoder, giving a theoretical 8 uOP total . but the reality is, this change brought about only a >10% increase int total throughput.

Whats important though is that a from a high level overview- A Zen Core is , an enhanced BD/PD Module but with a single, wider integer cluster. So when you look at total throughput of each.. i.e two threads, you can expect much less than the 40% performance delta between them, especially since you lose the benefits of keeping those integer execution dedicated (i.e the scheduling becomes more complex to fully utilize the same resources / use them as efficiently)